Communications of the ACM
Research Suggests Diversity as Key to Human-AI Collaboration
Researchers say diversity is a quantifiable proxy for human preference.
Credit: Bryan Mastergeorge
Before humans and artificial intelligence can team up as true partners, researchers must overcome a problem that corrodes cooperation: humans often do not like or trust their AI partners.
New research points to diversity as being a key parameter for making AI a better team player. MIT Lincoln Laboratory researchers found that training an AI model with mathematically "diverse" teammates improves its ability to collaborate with other AI.
In addition, independent work published by both Facebook and Google's DeepMind made diversity a part of system training and improved outcomes in human-AI collaborative games.
"The fact that we all converged on the same idea — that if you want to cooperate, you need to train in a diverse setting — sets the stage for the future work in cooperative AI," says Ross Allen, an AI technology researcher at MIT and co-author of the Lincoln Laboratory paper, which was presented at the 21st International Conference on Autonomous Agents and Multiagent Systems.
From MIT News
View Full Article
No entries found